Bolt.new Review: AI App Builder, Pricing & My Test Results

Hey, I'm Miles, and this is RankDude, where I test AI tools hands on and publish every number, so you can decide before you pay. Today it's Bolt.new, an AI app builder. You describe an app in a chat, and it builds the screens, the database, the log-in and the hosting. I tested it on a paid Pro plan, twenty-five dollars a month, with one brief for a Pilates booking app called Fitting Room. The short version is that Bolt.new built the app from one message in about twelve minutes, but the first version was broken. Five fix messages put that right, and twenty-three of my twenty-four checks passed. One of my two attacks got through, because a member could make themselves admin. All twelve changes worked, and the code downloads as a project you can run. My rating is four out of five. That's the average of eight tests, with nothing added for my verdict.

So, what is Bolt.new? You get a chat box, a live preview, a data panel and hosting. You say what you need, and the AI writes the screens, sets up the database, adds the log-in and hosts the app. After that, every change is another message. And Bolt.new charges in tokens, which the chat spends every time it builds or changes something. Pro gives ten million tokens a month. Payments and emails are not built in. It asks you for your own Stripe and email keys, so I left them out of my tests.

Every builder I review gets the same brief, word for word. Fitting Room is a booking and membership app for a small Pilates studio. It has a timetable, members who book, coaches who see only their own classes, an admin, and a ninety-five dollar five-class pass. I pasted the brief in one message. Bolt.new asked two design questions, which I skipped, and told me it needed my Stripe keys. About twelve minutes later, the first version was ready, using about one point three million tokens. I published it from the chat in under two minutes. Looking good was not working. The sign-in and sign-up buttons led to pages that do not exist, so every visitor went back to the home page. An admin page crashed. Booking could never succeed, and the admin and coach screens showed no bookings. Five fix messages took about zero point eight million tokens. After the last one, one hour and twenty-five minutes after my message, twenty-three of the twenty-four checks passed. On my scale, the build earns a four.

Now accuracy. I ran the twenty-four checks that need no keys of my own against the live, published app, by hand and with scripts, with one screenshot for each. I left out the six payment and email checks, because they need your own keys. The final app passed twenty-three of twenty-four. The one that failed is the password reset, whose link goes to localhost. All ten facts from the brief were correct. But Bolt.new added three things nobody asked for: a Most popular tag on the pass, Monday through Saturday wording, and stock photos. The timetable shows a weekly pattern with day tabs, not dated classes. Accuracy scores four.

Log-in, and the two attacks. Sign-up worked for members, a coach and the admin, and a session survived a reload. The first admin account is created with a one-time code that Bolt.new printed in the chat. The password reset email arrived in eleven seconds, but its link opened a page on my own computer, so a new password could never be set on the published app. Attack one. I'm member A, and I ask for other members' bookings, profiles, credits and orders. Blocked. I got only my own rows. Attack two. A member calls admin-only actions. Editing or deleting a class was refused. But one request that set my own role to admin was accepted, and the app then treated me as an admin. I put the role back afterwards. One of two attacks blocked, so access control scores three.

Next, the data. The contact form and the bookings wrote real records, and they were still there after a republish and a design change. In Bolt.new's own data panel, I viewed the classes table, filtered it, added a row, deleted it, and exported the table as a CSV and as a JSON file. After all twelve changes, the five classes were still there. The panel listed ten tables. All six checks passed. Five out of five.

Payments and emails. I did not test them. Bolt.new does not set up Stripe or an email service for you. It needs your own Stripe secret key, a webhook secret and an email key, and I gave it none. The six checks that depend on them are left out of the grade. Here is what I saw without keys. The Passes page said Stripe was being set up, so no checkout opened. The admin's email outbox read No emails sent yet after several bookings. To keep testing booking, I asked Bolt.new to give every member five free credits, for my testing only.

Then I made twelve changes, one message each. Things like a phone field on sign-up, a renamed role, new colours, an About page, a filter, a waiting list, a dark mode toggle, and a bug I planted on purpose. After each one, I ran a short check as a member, a coach and the admin, not all the checks again. All twelve finished with one message each, and nothing that already worked broke. One change, the class size, was only a data edit and made no new version. One check, the second email, could not be done, because the app sends no emails without keys. Each change took one to seven minutes. All twelve used about three point four million tokens, and dark mode alone used about zero point nine million. Five out of five.

Publishing. Bolt.new put the app online in under two minutes, from a chat message, with no extra cost. After each later change, you press an Update button in the Publish dialog. I did not connect my own web address, so I can't say how that goes. On Google's phone speed test, the home page scored seventy-nine, seventy-eight and seventy-seven out of a hundred, and the main content took about four seconds to appear. I only ran the home page, did not count layout faults at different widths, and did not test on a real phone. That earns a three.

Can you take the app with you? Yes. In the project menu, Download gave me a zip of sixty-six files on the second try, because the first attempt showed no file. It's a Vite and React project, with the database migrations and three server functions in a Supabase folder. I found no Bolt-specific code in it. It installed and built cleanly on my own computer in three seconds. To run it elsewhere, you need your own Supabase project, the migrations applied, the functions deployed and two values set. I didn't run it end to end, and I didn't read Bolt.new's terms, so I can't say who owns what you build. Export scores four.

Support and cancelling. Bolt.new's email support is for paid plans only, and its Help Center points free plans to Discord. The Help Center assistant failed browser verification, so I emailed my question to support on the twenty-sixth of October. I'll add the reply to the written review when it comes. Cancelling the Pro plan took four steps: Settings, Subscription and Tokens, Billing details, then Cancel subscription. Bolt.new warns that branding returns, private sharing switches off and banked tokens become inaccessible, then asks for a reason. The plan moves to Free after the twenty-fifth of November, and the app and projects stay. Support is not graded yet.

What did it cost to finish? In total I paid twenty-five dollars: one month of Pro, plus tax. The whole test used five point five million of the plan's ten million tokens. That's fifty-five percent of the month, about thirteen dollars seventy-five at the plan price. The first build used one point three million tokens, the five fixes zero point eight million, and the twelve changes three point four million. Running the app costs more than building it, because payments and emails need your own accounts, and their fees aren't in this number. The next plan up is fifty dollars a month, for twenty-six million tokens.

What do other users say? On Trustpilot, Bolt.new scores two point one out of five from two hundred thirteen reviews, which I read on the twenty-sixth of October. Trustpilot's summary of recent reviews lists the AI crashing, using too many tokens, struggling to fix its own mistakes, and slow support. My test is one app over two days. I did meet some of it: a crash, a booking button that never worked, and five fix messages. Support hasn't replied yet, so I can't judge it.

Pricing. Bolt.new has a free plan, but I didn't read its limits. Pro is twenty-five dollars a month, billed monthly, for ten million tokens, with your own domain and no Bolt branding. Teams is thirty dollars per member, and Enterprise is a custom contract. Paying for a year costs less. Finishing my app took about half of Pro's tokens, so Pro was enough to build and change it. Pricing scores four.

So, my verdict. Bolt.new is worth trying if you'll check what it builds, and can bring your own Stripe and email keys. It made a nice-looking app in twelve minutes, every later change worked in one message, and the code downloads as a project you can run yourself. The weak points are a first version with broken links, a crash and booking that couldn't work, and a member who could make themselves admin. I'd use it with a developer beside me. The eight graded tests average four point zero out of five, and my rating is four out of five. Payments and emails need your own keys and support awaits a reply, so neither is graded. Every measurement, every screenshot and every price is in the written review and the test results, linked below.